A META-ANALYSIS OF SECOND LANGUAGE CLOZE TESTING · PDF fileA META-ANALYSIS OF SECOND LANGUAGE...

Second Language Studies, 26(2), Spring 2008, pp. 103-133.

A META-ANALYSIS OF

SECOND LANGUAGE CLOZE TESTING RESEARCH

YUKIKO WATANABE & DENNIS KOYAMA

University of Hawai‘i at Mānoa

INTRODUCTION

Cloze procedure first appeared in 1953 when Wilson Taylor researched its effectiveness as a

procedure for estimating the readability of textbooks for school children in the United States. A

decade later, research began to appear on the usefulness of cloze for testing the reading

proficiency of native speakers of English (e.g., Bormuth, 1965, 1967; Crawford, 1970; Gallant,

1965; or Ruddell, 1964). In the 1960s and 1970s, a number of studies focusing on second

language learners emerged and the usefulness of cloze procedure as a measure of overall ESL

proficiency was examined (see Alderson 1978; Oller 1979; Cohen 1980 for summaries of this

early ESL research). Since then, the cloze procedure has gained popularity and is a commonly

used test in many language programs.

The literature on cloze testing now spans more than half a century of research and covers a

diverse range of topics. Researchers have manipulated an assortment of variables including, but

not limited to: deletion patterns (for example, every nth word, or rational deletion), starting point

of deletions (for example, after the first sentence, or prescriptively at the nth word), scoring

methods (such as exact answer or acceptable answer), text difficulty (using a readability index,

or the source of text; see, Brown, 1993), and the number of items—to name just five of the many

variables that have been researched.

With such a myriad of research topics both within and among studies, it not surprising that

the field of cloze test research has produced varying results. One noteworthy topic is that of

reliability. The reliability estimates for various cloze tests, over the years, have spanned the

reliability spectrum from both weak 0.31 (Brown, 1982) to very strong 0.95 (Brown 1978)

estimates. Many cloze studies (e.g., Alderson, 1979a, 1979b 1980; Bachman, 1985; Brown, 1980,

1983, 1984, 1988b, 1989, 1993, 1994, 1998, 2002; Brown, Yamashiro, & Ogane, 1999, 2001;

Conrad, 1970; Darnell, 1970; Hinofotis, 1980; Irvine, Atai, & Oller, 1974; Jonz, 1976; Mullen,

A META-ANALYSIS OF SECOND LANGUAGE CLOZE TESTING RESEARCH

104

1979; Oller 1972a, 1972b; Oller & Inal, 1971; Pike 1973; Revard, 1990; Stubbs & Tucker, 1974)

have investigated how different approaches to creating, scoring, and interpreting, cloze tests

could be used to maximize their reliability, validity, or both. This trend shows that much research

on cloze procedures has focused on cloze procedures as a test, whether a diagnostic test, a

placement test, a proficiency test, or otherwise. The lively research brewing on the crafting of

cloze tests, caused some researchers to argue about the constructs that cloze items were primarily

tapping. Some researchers argued that cloze tests assessed student abilities to handle clause or

sentence level grammar (e.g., Alderson, 1979a; Markham, 1985; Porter, 1983), while other

researchers were arguing that cloze items measure at the intersentential level (e.g., Bachman,

1985, Brown, 1983, 1994; Chavez-Oller, Chihara, Weaver, & Oller, 1985; Chihara, Oller,

Weaver, & Chavez-Oller, 1977; and Jonz, 1987, 1990).

Use of Meta-analysis for Informing Test Design

As mentioned above, cloze test studies have produced a variety of contradictory results. One

way to untangle these contradictions is to systematically analyze the characteristics of the

primary studies and identify moderating or confounding variables; this approach is often called

research-synthesis1. The accumulated information (e.g., effect sizes, descriptive statistics, and

reliability estimates) from primary studies will provide an extensive picture of the relationships

among variables—and in some cases—provide direction and the magnitude of the effects across

studies (i.e., meta-analysis). Given the popularity of the use of cloze tests for various purposes, it

is important for test designers to be aware of different ways to craft and tailor cloze tests, as well

as the causes of biased test construction2 that may lead to spurious results and irresponsible

decisions.

Numerous studies have accumulated over the last forty years in which second language

testing researchers have examined various cloze test characteristics, including what cloze tests

are measuring, under what conditions, and for what type of learners. In order to understand the

1 Meta-analysis and research-synthesis have been widely used across many fields, including but not limited to medicine, education, psychology, and more recently in second language acquisition research (see Norris & Ortega, 2006 for a comprehensive overview and various meta-analytic and research-synthesis studies on language learning and teaching). 2 The Association of Language Testers in Europe (1990) succinctly defines test bias concerns as follows: “A test or item can be considered to be biased if one particular section of the candidate population is advantaged or disadvantaged by some feature of the test or item which is not relevant to what is being measured” (p. 136).


105

accumulated knowledge of cloze procedures in second language testing research, the following

research questions were posed:

1. What are the test and learner characteristics in the primary cloze test studies to date?

2. How does the deletion pattern affect the reliability of the cloze results?

3. How do various scoring methods (exact, acceptable, and clozentropy) affect the reliability of

the cloze tests?

4. What is the strength of relationship between acceptable and exact scoring methods?

The following section describes the study identification procedures, the study inclusion and

exclusion criteria, and the coding system. The findings from the above research questions are

presented in two sections. The first section summarizes the results of the research synthesis,

identifying the characteristics of the study, the test, and the participants, while the second section

delineates the results of the meta-analysis of 33 studies.

METHOD

The Literature Search

We screened three literature databases, Education Resources Information Center (ERIC),

Linguistic and Language Behavior Abstracts (LLBA), and PsychInfo, to identify empirical

studies on cloze procedures since 1953, the year of Taylor’s seminal article. The following

search terms were utilized: “Cloze procedure” (descriptor) AND “language test” NOT “dyslex*,

disab*.” Since the cloze procedure is often utilized for testing learning disability (e.g., dyslexia),

we excluded studies that has “dyslex*” and “diab*” as keywords to avoid introducing additional

learner variables. The database search identified a total of 114 studies including 16 dissertations

(See Table 1). To assure an exhaustive search, we also manually searched Language Testing

Journal, locating two more studies. An additional 96 studies were discovered through footnote

chasing, resulting in total of 212 studies.


106

Table 1

Literature Database Search

Note. ( ) indicates the number of dissertations in addition to the journal articles.

The Inclusion and Exclusion Criteria

The following four criteria were set to determine whether the retrieved studies qualified for

inclusion in the research synthesis and meta-analysis:

1. The study is an empirical study published between 1957 and March, 2007.

2. The study participants are adult English as a second or foreign language learners. The scope

of the current study is strictly limited to English as a target language, since English cloze

tests are most studied among other languages. Because it is not clear whether cognitive

development affect cloze test results, we decided to only include adult learners and avoid

populations under 18 years of age.

3. The study used at least one cloze test. Here, we define cloze test as a test that has certain

words deleted from a passage. We did not include gap-filling test that has only one sentence

as a stem, such as section two in the computer-based TOEFL (also see example below).

Because of the storm and rough waves, it would be foolish to go out sailing today in a small ____.

a) automobile b) house c) boat d) beast

(Perkins, & Linnville, 1987, p. 128)

4. The study adequately described the cloze test employed in the study, so test characteristics

can be coded.

Studies were excluded from the analysis, if any of the following criteria were met:

1. The study was published as conference proceedings or in-house publication (working paper),

or was an unpublished manuscript (master’s thesis or dissertation).

2. The study employed Analysis of Variance, and did not report any descriptive statistics.

3. The cloze test was used just to determine learner characteristics (e.g., grouping learners into

high and low proficiency groups), and was not the main focus of the study.

LLBA ERIC PsychInfo Total without overlap

(Cloze procedure) AND (language test) NOT (dyslex, disab)

61 (3) 20 (0) 39 (13) 114 (16)


107

Based on the inclusion and exclusion criteria, 33 studies remained for the research-synthesis

and meta-analysis. After determining the number of studies to be included in the meta-analysis,

we set a coding scheme for extracting study information (research questions and study setting),

learner characteristics (number of participants, L1, and language proficiency), test characteristics

(text length, number of items, response type, deletion pattern, starting point of deletion, and

scoring criteria), and test results (descriptive statistics and reliability estimates). Table 2

summarizes the coding scheme we utilized. The interrater reliability for two raters on four

independently read research studies was 93.5%. Any inconsistencies and discrepancies were

resolved though discussions and the end agreement percent was 100%.

Table 2

Coding Scheme

Study information Learner characteristics Test characteristics Quantitative data

- Author(s) & publication year

- Research question(s) - Study setting (EFL or

ESL)

- Number of participants - First language - Age - English proficiency (researchers’ description on proficiency)

- Text length (total # of words) - Text difficulty (Flesch-Kincaid, etc.) - Response type (open-ended or

multiple choice) - Total number of items per passage. - Deletion pattern (nth word deletion,

rational deletion) - Starting point of deletion - Scoring criteria (exact, acceptable,

clozentropy)

- Descriptive statistics (mean, standard deviation, min, max)

- Reliability estimate (split half Spearman-Brown coefficient, KR-20, KR-21, Cronbach’s alpha)

RESULTS AND DISCUSSION

Research Synthesis on Study Characteristics

Learner characteristics. To identify the proficiency levels stated by the researchers, we

adapted the four proficiency categories suggested by Thomas (1994, 2006): impressionistic

judgement (IJ), institutional status (IS), in-house assessment (IH), and standardized test (ST). In

addition to these four classifications of proficiency, we added two others3, experience (EX), and

self-assessment (SA). Following Thomas’s guidelines, impressionistic judgement means the

author has suggested an second language (L2) ability based on the author’s impression, for

3 For studies that did not clearly articulate learner proficiency, we labelled them as ‘no mention’ (NO).


108

example, “The examinees were intermediate-level”. This label of ‘intermediate-level’ is not

usually qualified by any concrete evidence and is therefore considered impressionistic.

Institutional status is similar to impressionistic judgement in that there is no evidence that

indicates learner proficiency level, rather an association of ability and institutional status is made

by categorizing students, for example, “The students in this study were English majors”. This

sort of labelling is often combined with IJs, such as, “The examinees in this study were second

year English majors and had an intermediate-level of English”. In-house assessment was used to

identify studies that offered proficiency assessments based on institutional or in-class proficiency

exams, for example, “The students took a placement exam that consisted of an essay test.” The

standardized test classification was used when a researcher noted students’ abilities on widely

used standardized exam, such as, “The students had a range of 300 to 500 on the TOEFL exam.”

Students’ language proficiency was categorized as EX when the study interpreted their learning

history through the students’ experiences, for example, “All the examinees had spent at least 2

years abroad.” Self-assessment was used when researchers collect information from the students

through interviews or questionnaires. Students may have had to answer a question item on a

survey like, “At which level would you rank your English ability: advanced, intermediate, or

beginning.”

Table 3 describes and summarizes the learner characteristics. This description includes

details about the number of participant subgroups studied, study settings, (ESL, EFL), L1s,

proficiency assessments of the subjects, and the number of participants. Across the 33 studies

analyzed in this meta-synthesis, 17 studies were in ESL settings and 15 studies in EFL settings.

One study (Oller & Inal, 1971) was conducted both in ESL and EFL settings. In total, there were

44 independent distinct subgroups across the 33 studies. By distinct, we mean studies that

collected data at clearly different times and administrations (e.g., fall term students, spring term

students, etc.), or studies that collected data on wholly separate populations in different countries,

institutions, or locations (e.g., data collected in Japan compared with data collected in Peru).

Seven studies had multiple sub-groups within one study. Within the 44 sub-groups, there were 23

sub-groups of ESL learners and 21 sub-groups of EFL learners.

The most common L1 classification was “various” with ten studies followed by: Japanese

with five sub-groups, Chinese and Iranian with three sub-groups each, and Spanish had two sub-

groups. Other L1s (in alphabetical order) included: Farsi, German, Indo-European, other S.E.


109

Asian, Turkish, and Vietnamese. The participant size per sub-group varied greatly, ranging from

as little as six to as many as 14,613. The average participant group size was 647.86, with a

majority of the sub-groups (n = 28) with 101 to 500 participants, followed by studies with 100

people or less (n = 10), and finally, studies with more than 500 participants (n = 6).

Table 3

Learner Characteristics

Context N Participant subgroups N4 L1a N Proficiency b n Participants N ESL 17 Studies with 1 subgroup 26 various 10 NO 8 101-500 28 EFL 15 Studies with over 1 subgroup 7 Japanese 5 IH 5 20 (Min)-100 10 ESL/EFL 1 Chinese 3 ST 3 500-14613 (Max) 6 Iranian 3 NO/EX 3 Spanish 2 IS 2 OtherC 6 EX 2 IS/ST 2 (Average) (647.86) IS/EX/ST 1

IH/EX 1 IH/EX/SA 1 IH/IJ 1 IJ 1 IJ/EX 1 a Number of L1 subgroups across all studies that indicated by the authors. The following L1s were reported once: Farsi, German, Indo-European, Other S.E. Asian, Turkish, Vietnamese. b Proficiency abbreviations are: institutional status (IS), in-house assessment (IH), standardized test (ST), impressionistic judgment (IJ), experience (EX), self-assessment (SA), no mention (NO). Test characteristics. In addition to the characteristics of students, the characteristics of each

independent cloze test were noted. In particular, the number of studies that reported test level

characteristics such as: response type (e.g., open-ended or multiple-choice), deletion pattern (e.g.,

nth deletion, or rational), scoring method (exact, acceptable, or clozentropy), text length (i.e.,

total number of words in the test), total items (i.e., the number of cloze items), and readability

indexes (e.g., Flesch, Flesch-Kincaid, Gunny Fog, or Fry).

There are essentially two response types for cloze tests, open-ended (OE) and multiple-

choice (MC). These response types are inextricable linked to scoring methods. For an open-

ended cloze test, there are three ways to score the test: exact scoring, (EX; i.e., the response must

4 N indicates the number of studies, whereas n represents distinct sub-groups within a study. For example, if three classrooms take the same cloze test, these are not distinct groups for research purposes and are counted as one group with an aggregated population size. Conversely, a study that collects cloze test results from separate administrations (i.e., to different students, such as an entrance exam across two separate terms) is considered a multi-group study.


110

match the original phrasing of the given blank), acceptable scoring (AC; i.e., the response must

be grammatically and/or semantically acceptable response, often judged by a native speaker),

and clozentropy scoring, which is a scoring system developed by Darnell (1970). For

clozentropy scoring (CE), a corpus of answers is developed and the frequency of each answer on

the corpus is calculated. Then, this information is cross-checked by administering the same cloze

test to native speakers and compiling those responses. The answers are, finally, weighted

according to a logarithm of the frequency of each response. Multiple-choice responses, however,

necessarily must be EX answer method, as it makes little sense to construct a test where the

responses include a list of acceptable answers. Notably, though, the matter of deletion is

divorced from the response type, and therefore is not affected by the method of scoring.

The primary forms of deletion are nth deletion pattern and the rational deletion. The nth

deletion is a pseudo-random form of deletion wherein the test creator selects a number say, six,

and eliminates every sixth word from the passage. This means, however, the lower the number,

the greater the likelihood that more than one word will be deleted from a single sentence.

However, some authors account for this issue and consider other mitigating factors when

deleting a word. For example, Kobayashi (2002) used an nth word deletion, but avoided deleting

proper nouns and numbers. If the 13th word was a proper noun or number, the previous word (i.e.,

the 12th word) was selected for deletion. The last form of deletion, rational deletion, is patterned

targeting specific linguistic forms or words. For example, Oller and Inal (1971) deleted every

other preposition in the construction of their test.

Table 4 shows the count of test characteristics for independent tests across all of the studies

(k = 157). 5 The most frequently used response type was open-ended with 139 tokens, and next

was multiple-choice (k = 13). One unique study by Hinofotis and Snow (1980) had two response

types within each of two independent tests. Their article indicates that two test forms (Form A

and Form B) had 50 items for each passage and each form was constructed so that in Form A

“the first 25 items were MC and the second 25 were open-ended. On Form B the order was

reversed” (Hinofotis & Snow, 1980, p. 130). This means one fifty-item passage had two

response types. The most common scoring method was EX with 75 tests, next was the combined

methods of EX and AC (k = 50) which had 50 independent tests, and the third most common

5 k denotes the number of independent test.


111

format of scoring was AC (k = 15). Thirteen authors chose not to state the method of scoring, and

one study (Brown, 1993) used all three methods of scoring, AC, EX, and CE.

There were 15 different styles of deletion, and 12th word deletion was the most prevalent

with 55 independent tests utilizing this format. The second most common deletion pattern, 7th

word, had 25 cases, followed by rational deletion (20 cases). One unique study, Brown (1988),

used a range of deletion patterns where words were deleted in such a way that blanks were never

closer than every fifth word and no farther apart than every ninth word. Another study (Jonz,

1976) did not state the deletion pattern.

Table 4

Tally of Test Characteristics across Independent Tests

Response type k Scoring k Deletion pattern k OE 143 EX 75 12tha

55 MC 13 EX/ACb 50 7th 25 OE & MC 2 AC 15 Rational 20

No mention 12 10th 13 EX/CE 3 11th 8 AC/CE/EX 1 13th 8 6th 7 8th 6 15th 3 5th 3 9th 3 16th 1 18th 1 4th 1 5w - 9w 1 No mention 1 aBrown (1993) had 50 independent tests using 12th deletion, and one study (Bachman, 1982) noted the deletion pattern was on average 12th. bBrown (1993) had 50 independent tests using this system of scoring. A closer examination of the studies reporting on the make-up of the test is summarized in

Table 5. This table identifies the total number of tests that state the length of text and total

number of items on the test. Descriptive statistics particular to these characteristics is also given.

There were a total of 69 tests that gave the length of the cloze passage (or the length of the

passage was calculated from the test in the appendix of the article, as in the case of Chapelle

(1988) and Chapelle and Abraham (1990)), additionally, 137 tests reported the number of total

cloze items. The range of text length was 125 words to 750 words with an average of


112

approximately 374 words per passage, and the standard deviation was 126.53 words. Of the 137

studies that reported the number of cloze items, the range was 15 to 80, and the average number

of items was 34.34 with a standard deviation of roughly ten cloze test items. Although this study

set out to list the common readability indexes across the studies, very few authors offered a

readability estimate of their cloze passage. However, of the studies that did give a readability

index, the most common readability given was the Flesch (n = 4), followed by the Fry index (n =

3). The Gunning-Fog and the Flesch-Kincaid were noted only once. One study by Brown,

Yamashiro, and Ogane (2001) rated the passage as coming from an introductory reading text,

and another study (Mullen 1979) described the level given in the manual from which the text was

extracted as approximately 7th and 12th grades.

Table 5

Tally of Text Characteristics across Independent Tests

Number of test reporting Range M SD Text length 69 125 - 750 373.71 126.53 Total items 137 15 - 80 34.34 10.28

Reliability

The concept of reliability estimates as a measure of a given tests consistency in measuring a

particular construct (or, multiple constructs) has a long history. The most common formulas for

estimating test reliability are: K-R20, K-R21, Spearman-Brown, and Cronbach alpha. Table 6

shows the break down of reliability estimates given. By far the most common reliability estimate

given was K-R20 with ten studies using it solely, followed by three studies that used K-R21 and

three other studies that used both Spearman-Brown (S-B) and Cronbach (α) estimates of test

reliability. Stand-alone estimates of reliability were limited to two studies that used α, and

another study that used S-B only. Five other studies used various combinations of reliability to

give the reader a broader perspective of the test’s reliability. For example, one study used both

K-R21 and α while another study reported all four forms of test reliability estimates. The most

frequently used estimates were as follows: K-R20, (used alone or with another estimate in 13

studies), Spearman-Brown and Cronbach α (which were used alone or with other estimates in

seven studies), and K-R21 (used alone or with another estimate in four studies).


113

Table 6

Reliability Estimates Used by Studies

a One unique study reported Guttman’s lower bounds to reliability estimates, and one study offered no name for the reliability estimate.

Meta-analysis on Cloze Test Reliability Estimates

A total of three statistical analyses were conducted analyzing the reliability estimate by

scoring system and by deletion pattern. When performing multiple statistical tests, a family-wise

error rate of p < 0.05 is maintained within a study to reduce false positives (i.e., Type I error)

among a class (family) of tests, under the null hypothesis. Since the current study performed

three statistical tests, an a priori decision was made to report data for a dependent variable as

statistically significant only when multivariate F values and the t-test value were significant at p

< 0.017 (p < 0.05 divided by three, as dictated by the Bonferonni adjustment).

Scoring system and reliability. A total of 223 reliability estimates (across 24 studies)

associated with each scoring method were analyzed. From the descriptive statistics (see Table 8),

the values for exact scoring were more widely dispersed than those for any other methods. This

can be attributed to the few studies that used tests with very low reliability. For example,

Farhady and Keramati (1996) used a structure-driven deletion cloze test which produced K-R21

= 0.14, 0.23, and 0.40. Brown’s (1983) study also contributed to the large variability in the

overall exact scoring method, since his study found nine cloze tests that produced reliability

estimate equal to or less than α = 0.50.

A one-way analysis of variance (ANOVA) was conducted with reliability as a dependent

variable and scoring method (acceptable, exact, and clozentropy) as an independent variable to

investigate whether there were any significant differences in reliability among scoring methods.

Across 223 cases of reliability estimates extracted from 24 studies, one way ANOVA revealed

that the scoring system makes a difference in the reliability results, F(2, 220) = 16.06, p = 0.001.

Type(s) of reliability specified a N

K-R20 11 Cronbach alpha ( α ) 3 S-B & K-R20 3 Spearman-Brown (S-B) & Cronbach’s alpha 2 K-R21 2 K-R21 & Cronbach’s alpha 2 S-B, K-R20, Split-half 1


114

A post hoc Tukey HSD test indicated that the exact scoring method produces the lowest reliable

among the three scoring methods, and that the acceptable and clozentropy differences are

statistically non-significant (see Table 9). However, these results have to be interpreted with

caution, since there were a number of cases where clozentropy was less than other two scoring

methods. As can be seen in Figure 1, the error associated around the mean in the 95% confidence

interval was larger in magnitude for clozentropy than EX and AC scoring methods.

Table 8

Descriptive Statistics of the Reliability Estimates across Scoring Methods

k M SD Min Max Acceptable 97 0.74 0.12 0.60 0.97 Exact 122 0.64 0.16 0.14 0.99 Clozentropy 4 0.86 0.06 0.78 0.91

Table 9

Post-hoc Tukey HSD on Scoring Methods

Scoring comparison Mean difference p AC-EX 0.10 0.001* CZ-EX 0.22 0.008* AC-CZ -0.12 0.241

*p < 0.017


115

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

0.90

1.00R

elia

bili

ty e

stim

ate

.

95% C.I. lower bound 0.72 0.61 0.77

Mean reliability 0.74 0.64 0.86

95% C.I. upper bound 0.77 0.67 0.95

Acceptable (k = 97) Exact (k = 122) Clozentropy (k = 4)

Figure 1. 95% Confidence Interval Around Mean Reliability for Each Scoring Method A more fine-tuned analysis was performed with studies that compared different scoring

systems (N = 12). However, since only two studies (Brown, 1980; Pike, 1979) investigated the

reliability difference between clozentropy and other scoring methods, we included only 10

studies that utilized both the exact and acceptable scoring methods when scoring the same cloze

test.

The test characteristics (e.g., total number of items, deletion pattern), number of

participants for each cloze test, descriptive statistics, and the reliability estimates are summarized

in Table 10. The average difference between EX and AC scoring methods was 0.068, the range

was from -0.11 to 0.43. A paired t-test was used to determine if the scoring systems made a

difference in test’s reliability. The analysis revealed that the AC scoring system produced greater

reliability than the EX scoring system with statistical significance, t(37) = 4.10, p = 0.001. This

result is reasonable, given that the AC scoring will result in larger means and standard deviation

(Overall EX: M = 14.86, SD = 3.78; AC: M = 22.25, SD = 4.61). Across 38 independent tests,

the overall relationship between the two scoring methods and reliability was r = 0.793 (p =

0.001). Therefore, 63 percent of the variance in the reliability in one scoring method (EX or AC)

can be predicted by the other scoring method (EX or AC). In sum, the results in this section


116

indicated that AC scoring is a more reliable scoring method than exact scoring, which is in line

with the findings of Brown (1980, 1983).


117

Table 10

Reliability Difference between Exact and Acceptable Scoring

Deletion pattern and reliability. Two hundred twenty-three cases of reliability estimates

extracted from 24 studies, were classified based on the deletion pattern of the cloze test. Due to

Author Test ID # items Delet. Pattern

N Exact Acceptable (sem/gram) Reliability (AC-EX)

M SD Relia-bility M SD

Relia-bility

Brown (1980) 50 7th 55 15.00 8.54 0.90 25.58 12.45 0.95 -0.04Brown (1983) 50 7th 66 21.21 4.31 0.61 32.99 4.98 0.67 -0.06Brown, Yamashiro, & Ogane (2001)

Original 30 11th 40 1.58 2.26 0.76 1.80 2.33 0.74 0.02Orig+1R 30 12th 38 2.53 2.27 0.66 2.92 2.45 0.66 0.00Orig+2R 30 13th 39 0.72 1.32 0.64 1.64 1.71 0.53 0.11Orig-1L 30 10th 38 1.53 1.67 0.56 2.13 1.76 0.48 0.08Orig-2L 30 9th 38 3.50 3.22 0.79 4.05 3.32 0.77 0.02

Hinofotis (1980) 50 7th 107 11.90 2.08 0.61 15.30 7.30 0.85 -0.24Jafarpur (1996) 25 6 th 325 8.96 3.60 0.75 15.92 5.60 0.88 -0.13Jonz (1987) b

Reg_fam 50 15th 53 27.28 3.38 0.89 38.66 3.83 0.89 0.00Reg_unfam 50 15th 59 23.22 2.42 0.70 38.64 3.39 0.89 -0.19Coh_famc 50 Rational 53 31.42 3.95 0.89 41.58 3.07 0.89 0.00Coh_unfamc 50 Rational 65 16.52 2.98 0.79 34.22 4.40 0.92 -0.13

Klein-Braley (1983)

1 34 9th 23 16.34 3.48 0.62 25.95 3.46 0.63 0.012 46 7th 23 24.74 4.56 0.67 31.44 4.76 0.77 0.103 30 10th 45 14.22 2.50 0.15 21.80 3.15 0.58 0.434 43 8th 45 15.02 3.47 0.46 31.53 4.29 0.60 0.145 35 10th 31 15.97 3.51 0.58 26.58 4.68 0.77 0.196 50 6th 31 21.71 3.98 0.63 40.58 4.69 0.73 0.107 35 10th 29 12.00 2.92 0.41 23.35 3.74 0.59 0.188 50 6th 29 23.62 4.37 0.67 39.41 4.95 0.74 0.079 40 7th 23 20.64 4.21 0.71 31.73 3.94 0.65 -0.0610 30 10th 23 16.41 3.31 0.53 23.36 3.17 0.61 0.0811 40 7th 53 23.07 4.02 0.74 30.96 4.13 0.70 -0.0412 30 10th 53 20.52 2.46 0.41 24.11 2.30 0.50 0.09

Kobayashi (2002)

1 25 13th 63 7.94 2.77 0.62 11.50 3.75 0.72 -0.12 25 13th 66 6.91 3.32 0.72 10.70 4.65 0.80 -0.083 25 13th 61 8.31 4.56 0.82 9.36 4.92 0.84 -0.024 25 13th 65 5.46 3.36 0.73 8.54 4.30 0.79 -0.065 25 13th 63 7.89 3.66 0.75 9.68 4.43 0.80 -0.056 25 13th 66 7.35 3.39 0.73 9.15 4.23 0.80 -0.077 25 13th 61 7.77 3.58 0.73 9.93 4.40 0.80 -0.078 25 13th 65 5.29 2.65 0.62 7.33 3.73 0.77 -0.15

Mullen (1979) Easy 50 10th 154 --- --- 0.83 --- --- 0.91 -0.08Hard 46 10th 154 --- --- 0.73 --- --- 0.87 -0.14

Oller (1972) 1 50 7th 132 31.74 6.00 0.99 42.99 6.59 0.96 0.032 50 7th 134 33.85 8.77 0.98 41.52 9.64 0.93 0.053 50 7th 129 22.91 9.17 0.98 34.21 11.49 0.95 0.03


118

the small number of cases, fourth to sixth deletion patterns as well as 15th to 18th deletions were

treated as one group. Rational deletion and tailored deletion7 were also separated from random

deletion method. Table 11 details the descriptive statistics of the aggregated reliability for each

deletion pattern8. The most frequently used deletion pattern—across studies that reported

reliability—was seventh deletion (N = 10), followed by rational deletion (N = 5). The 12th

deletion had the largest number of independent tests with reliability information (k = 54), a

majority (k = 50) was contributed by a single study, Brown (1993). The rational deletion had the

highest mean reliability (M = 0.80) in opposition to random deletion patterns. The least reliable

deletion pattern was the eighth deletion method.

In further analysis, a one-way ANOVA was carried out to assess whether the deletion pattern

made a difference in reliability scores. The results revealed an overall statistical significance at p

< 0.017 level for the effect of deletion pattern on reliability, F(10, 174) = 4.921, p = 0.001. The

post hoc Tukey HSD test demonstrated only few statistically significant differences among

deletion patterns at p < 0.017 level. The seventh word random deletion test obtained greater

reliability than the tenth and twelfth word random deletion tests (see Table 12). Figure 2 depicts

the 95% confidence interval around the mean for each deletion pattern.

7 Tailored items have been deleted due to some tailoring criteria, such as item facility and item difficulty calculations. 8 The deletion patterns discussed here are deletions noted in research that clearly states reliability estimates per test.


119

Table 11

Descriptive Statistics of Reliability Scores for Each Deletion Pattern

Deletion pattern M SD Min Max 4-6th (k = 9; 4 studies) 0.75 0.09 0.63 0.88 7th (k = 28; 10 studies) 0.80 0.16 0.27 0.99 8th (k = 3; 2 studies) 0.54 0.07 0.46 0.60 9th (k = 5; 3 studies) 0.68 0.09 0.59 0.79 10th (k = 22; 4 studies) 0.65 0.20 0.15 0.91 11th (k = 9; 4 studies) 0.78 0.05 0.74 0.88 12th (k = 54; 3 studies) 0.63 0.13 0.28 0.82 13th (k = 26; 2 studies) 0.75 0.08 0.53 0.84 15-18th (k = 11; 2 studies) 0.69 0.29 0.14 0.92 Rational (k = 15; 5 studies) 0.80 0.08 0.61 0.95

Table 12

Post Hoc Tukey Test on Deletion Pattern (only statistically significant results)

Deletion pattern Mean Difference p

7th-12th 0.17 0.000* 7th-10th 0.15 0.014* Rational-12th -0.16 0.007*

*p < 0.017

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

0.90

1.00

Rel

iabi

lity

estim

ate

.

95% C.I. upper bound 0.82 0.87 0.72 0.80 0.74 0.82 0.67 0.78 0.89 0.84

Mean reliability 0.75 0.80 0.54 0.68 0.65 0.78 0.63 0.75 0.69 0.80

95% C.I. lower bound 0.68 0.74 0.36 0.56 0.56 0.75 0.60 0.72 0.50 0.75

4-6th (k= 9; 4

studies)

7th (k = 28; 10 studies)



10th (k = 22; 4

studies)

11th (k = 9; 4

studies)

12th (k = 54; 3

studies)

13th (k = 26; 2

studies)

15th (k = 11; 2

studies)

Rational (k = 15; 5 studies)

Figure 2. 95% Confidence Interval Around Mean Reliability for Each Deletion Pattern


120

DISCUSSION AND CONCLUSION

Through the consistent coding of substantive and methodological features across 33 studies,

the research synthesis and meta-analysis provided comprehensive profiles of cloze test research

and revealed what test-developers need to consider when creating and implementing a cloze test.

This section provides a discussion and suggestions for improving the reporting of research on

cloze testing, including: learner and test characteristics and the use reliability estimates.

An equal number of studies addressed both the ESL and EFL populations. The identification

of the population types (i.e., ESL, EFL) was not as much an issue as the clear identification of

the learner proficiency levels. For example any number of issues and concerns can be raised with

any of the statements of learner proficiency used across cloze test studies. Issues like, a

researcher who may deem their participants as ‘intermediate’ allows the reader to determine the

amount and quality of the students’ English abilities. This can be very difficult to do, especially

in the EFL based assessments as countries vary in the level and amount of English instruction.

This can be likened to judging the intensity a spicy dish on a menu in a restaurant that uses

pictures of chilli peppers to indicate the spiciness of the meal. Simply put, learner proficiency

statements seem to lack consistency across studies that are interpretable outside of the primary

investigators’ frames of reference.

In addition to a lack of consistency in identifying learner characteristics, many studies also

lacked detailed descriptions of test designs, scoring information, descriptive statistics, and

reliability estimates for the tests. Recall that among 158 independent cloze tests the OE response

was the most popular construction for cloze tests, and those responses were scored mostly with

the EX scoring method (see Table 4). This preference of style and scoring methods may be a

result of practicality involved in the ease of test construction and scoring. However, it is

noteworthy to highlight that AC scoring can be perceived as a much fairer way to score cloze

tests rather than EX scoring methods that only accept answers that were used by the author of the

original text. Finally, the method of CE may have pragmatic limitations to its construction and

implementation. This pragmatic concern stems from the underlying reason behind utilizing

clozentropy, which measures and weighs the linguistic ability of the test-takers to the target norm

(i.e., native speakers). The intrinsic focus on native speaker-like answers may call into question


121

the reasonableness of expecting non-native speakers to produce similar linguistic frequency

patterns as to those produced by native speakers.

Ultimately, choosing which scoring system to adopt is the decision of the test administrators.

For they are the ones who have to consider the purpose of the test, feasibility of creating and

utilizing a corpus of acceptable answers, errors that may accumulate during the scoring

procedures, and any other practical implementation issues.

Another important test design feature is the deletion pattern, which may be decided for any

number of reasons. Across the 33 studies, the seventh deletion pattern was the most common

deletion patterns (N = 14, k = 25), followed by rational deletions with seven studies using

rational deletions (k = 20). Just these two deletion patterns account for approximately 29 percent

of the studies analyzed and many of these studies did not cite the text length and total number of

items. These characteristics are invaluable to other testing researchers (not to mention test

construction artists), yet they frequently go unaccounted for. In addition to providing these

important test details it will also be necessary to state, when using rational deletion, the range

between deletions (i.e., the minimum and maximum distances between cloze items). In addition

to the reporting of these test description details, it would also be useful to always append the

actual test to the research report.

In this meta-anlaysis, we acknowledge that the aggregation and categorization of reliability

across different tests and studies by scoring pattern or test design pattern alone may introduce

other possible reasons for the variability of the test scores. Though we need to interpret the

results with extreme caution, we found that among scoring methods, clozentropy, was the least

used methodology but obtained the largest reliability estimate among the scoring methods. More

studies are needed that specifically look at the relationship between clozentropy and other

scoring methods within one study, so as to examine whether the differences in reliability are

related to other test features.

The deletion pattern comparison in terms of reliability showed that the seventh word deletion

and rational deletion cloze tests are most reliable. Despite the popularity of the seventh word

deletion pattern, more empirical studies are needed comparing different deletion patterns in order

to draw a more certain conclusion as to which deletion pattern is optimal.

Moving onto a discussion of reliability brings to the fore a very important, and fundamental

question: What assumptions do researchers make using different reliability estimates (K-R20, K-


122

R21, Spearman-Brown split-half reliability coefficient, and Cronbach alpha), for estimating

cloze test reliabilities? The presence of various reliability estimates across the studies in the

meta-analysis studies ignored a need to better understand what is actually going on with

reliability estimates as applied to cloze testing.

On the one hand, the K-R20 formula requires the use of three pieces of information, the

number of items, the standard deviation, and (perhaps most importantly) the average variance of

the items. On the other hand, the K-R21 formula requires similar types of test information, such

as the number of test items, the K-R21 also requires the average of the item scores. The key

difference, then, between the two formulas lays in the assumptions underpinning them. For the

K-R21 one assumption is “that the matrix of inter-item correlations has a rank of one, that these

correlations are equal, and that all items have the same difficulty” (Kuder & Richardson, 1937,

emphasis added)—an assumption that is probably never met in any reasonable way with cloze

test were item difficulty values often vary from .00 to 1.00. Thus the difference between K-R20

and K-R21 is clear: K-R 20 calculations estimate reliability with the use of the average of item

variances, whereas K-R21 uses the average of item difficulty (Kuder & Richardson, 1937, pp.

158-159). Therefore, the K-R21 formula may be inferior to the K-R20 formula when calculating

the reliability estimate for a given cloze test.

The issue of item difficulty is not limited to the discussion of the K-R21 formula. Item

difficulty should also be considered if the test’s reliability estimates are to be calculated with the

split-half Spearman-Brown coefficient because the difficulty of items in each of the halves of the

split test will necessarily affect the reliability estimate given by using the split-half method. That

is to say, an uneven split of difficult items and less difficult items may cloud the results of the

calculation, or boost reliability. As a consequence, extra steps to ensure an equilibrium between

test halves is necessary to produce fair and trustworthy estimates of reliability using the

Spearman-Brown split-half coefficient. Therefore, unless these extra steps are taken the

Spearman-Brown split-half coefficient may be a less desirable estimate of reliability when

compared to the K-R20 formula.

Cronbach alpha estimates how well a set of items (or variables) measures a single

unidimensional latent construct. When data have a multidimensional structure, Cronbach

alpha will usually be low. Since Cronbach alpha utilizes the Pearson product moment

correlation coefficient, it has been argued (see Farhardy, 1983) that Cronbach alpha, when used


123

on cloze test items, violates the assumption of item independence. Cziko (1978) and Brown

(1983) found cloze items to be dependent on previous context, thus we suggest that

Cronbach alpha may not be a theoretically appropriate reliability estimate to be used for

cloze testing.

Considering the discussion on reliability estimates, it may be credible to assert that the

K-R20 is, perhaps, the most trustworthy reliability estimate for cloze testing research.

Further research and investigation of this claim is necessary, including how item

interdependence may affect test reliability.

This research synthesis and meta-analysis highlights the gaps and insights of cloze testing

research; in sum, it is our hope that this study sheds light on the enterprise of cloze testing

research and test construction.


124

APPENDIX A

LEARNER CHARACTERISTICS

Author Setting Proficiency Context Subgroup N L1 Test name Range Judgmenta Abraham & Chapelle (1992) ESL 178 TOEFL 500+ ST/IS Alderson, C. (1979) ESL 360 NO Bachman (1982) ESL Fall 316 Various NO/IS ESL Spring 102 Various Bachman (1985) ESL incoming student 146 NO/IS/EX

entering full-time study 605

continuing study 61 Bensoussan & Ramaz (1984) Exp2 EFL University Freshman 691 NO/EX Experiment 3 EFL Entrance Examinees 14,613 NO Experiment 4 EFL Entrance Examinees 354 NO Bensoussan (1984) EFL Spring 2075 NO/EX EFL Summer 1420 NO/EX Briere & Hinofotis (1979) ESL UCLA 374 Various NO ESL USC 208

ESL SIU 107 Brown (1980) EFL 112 Brown (1983) EFL 125 Chinese TOEFL 390-590 ST Brown (1988) EFL 89 Chinese TOEFL 369-499b ST Brown (1993) EFL 2,298 Japanese NO/EX Brown, Yamashiro &Ogane (2001) EFL 144 Japanese IH/EX/SA Chapelle (1988) ESL 66 essay IS/IH Chapelle & Abraham (1990) ESL 201 TOEFL 500+ ST/IS Farhady & Keramati (1996) EFL 403 Iranian IS Flahive (1980) ESL 20 Various TOEFL 437-568 ST/IH Hinofotis (1980) ESL 107 Various IJ/IH Hinofotis & Snow (1980) ESL 66 Various NO Ilyin, Spurling, & Seymour (1987) ESL 257 IS/IJ Irvine, Atai, & Oller (1974) EFL 159 Farsi NO Jafarpur (1996) EFL 325 Iranian NO/EX Jafarpur (1995) EFL 325 Iranian NO/IS Jonz (1976) ESL 33 Various NO Jonz (1987) ESL Regular/cohesive &

familiar/unfamiliar 230 TOEFLc

(n = 100) IS/(ST) d

Jonz (1991) ESL Familiar/unfamiliar 238c TOEFLc

(n = 158) IS/(ST) d

Klein-Braley (1983) EFL 204 German EX, IJ Kobayashi (2002) EFL 255 Japanese NO/EX Mackay & Williamson (1979) EFL 22 NO Mullen (1979)e ESL 154 IS Oller & Inal (1971) ESL Winter 1970 110 Various ST/IH/IS EFL EFL Turkish 53 Turkish NO/EX/IH


125

Oller (1972) ESL University students 398 Various NO Pike (1979) EFL Peru 95 Spanish SA EFL Chile 143 Spanish SA EFL Japanese 192 Japanese SA Wilson (1980) ESL 72 Various IJ NOTE. a Proficiency abbreviations are: Institutional status (IS), in-house assessment (IH), standardized test (ST),

impressionistic judgement (IJ), no mention (NO), experience(EX), self-assessment (SA). b This is an estimate provided by the author, who clearly stated these numbers were estimated

c The number of students who reported TOEFL scores are not available for total N size. d The information on standardized test was only available for approximately half of the students. e Mullen (1979) stated TOEFL as a measure of proficiency for 80 participants; however, no specific scores were provided.


126

APPENDIX B

TEST CHARACTERISTICS

Author ID Text

length

Read-ability index

Readability value

Response type

Total items

Deletion pattern Scoringa

Abraham & Chapelle (1992) 1 OE 35 11th AC 2 OE 35 Rational AC 3 MC 35 Rational EX

Alderson (1979) 1 Teacher Difficult OE 6th EX, AC 2 Medium OE 6th EX, AC 3 Easy OE 6th EX, AC 4 Difficult OE 8th EX, AC 5 Medium OE 8th EX, AC 6 Easy OE 8th EX, AC 7 Difficult OE 10th EX, AC 8 Medium OE 10th EX, AC 9 Easy OE 10th EX, AC 10 Difficult OE 12th EX, AC 11 Medium OE 12th EX, AC 12 Easy OE 12th EX, AC

Bachman (1982) 1 365 OE 30 Ave.12th AC Bachman (1985) 1 330 OE 30 Rational AC 2 330 OE 30 11th AC Bensoussan (1984) 1 313 OE 26 7th AC Bensoussan & Ramaz (1984) Experiment 2

1 300 MC 21 Rational EX 2 300 MC 28 Rational EX 3 300 MC 24 Rational EX 4 300 MC 20 Rational EX

Experiment 3

1 300 MC 15 Rational EX 2 300 MC 17 Rational EX

Experiment 4 1 300 MC 41 Rational EX 2 300 MC 28 Rational EX 3 300 MC 28 Rational EX 4 313 OE 24 Rational AC

Briere & Hinofotis (1979) 1 427 Flesch 69.3 OE 50 7th EX Brown (1980) 1 399 OE 50 7th EX, AC, CE 2 399 MC 50 7th EX Brown (1983) 1 399 Fry Abt 8th grade OE 50 7th EX, AC 2 399 OE 51 8th EX, AC 3 399 OE 52 9th EX, AC Brown (1988) 1 399 OE 50 7th EX 2 399 OE 50 5th – 9th EX Brown (1993) 1 F-K 9.6 OE 30 12th EX 2 F-K 13.5 OE 30 12th EX 3 F-K 4.8 OE 30 12th EX 4 F-K 7.6 OE 30 12th EX 5 F-K 13.9 OE 30 12th EX 6 F-K 7 OE 30 12th EX 7 F-K 9.9 OE 30 12th EX 8 F-K 11.2 OE 30 12th EX 9 F-K 15.3 OE 30 12th EX 10 F-K 15.2 OE 30 12th EX 11 F-K 5 OE 30 12th EX 12 F-K 11 OE 30 12th EX


127

13 F-K 12.1 OE 30 12th EX 14 F-K 8.5 OE 30 12th EX 15 F-K 12 OE 30 12th EX 16 F-K 13 OE 30 12th EX 17 F-K 20.4 OE 30 12th EX 18 F-K 12.7 OE 30 12th EX 19 F-K 10.2 OE 30 12th EX 20 F-K 10.8 OE 30 12th EX 21 F-K 7.5 OE 30 12th EX 22 F-K 10.8 OE 30 12th EX 23 F-K 13.9 OE 30 12th EX 24 F-K 13.1 OE 30 12th EX 25 F-K 10.2 OE 30 12th EX 26 F-K 16.6 OE 30 12th EX 27 F-K 10 OE 30 12th EX 28 F-K 14.4 OE 30 12th EX 29 F-K 16 OE 30 12th EX 30 F-K 6.5 OE 30 12th EX 31 F-K 11.6 OE 30 12th EX 32 F-K 9.6 OE 30 12th EX 33 F-K 16.3 OE 30 12th EX 34 F-K 12.8 OE 30 12th EX 35 F-K 4.8 OE 30 12th EX 36 F-K 11.3 OE 30 12th EX 37 F-K 8.6 OE 30 12th EX 38 F-K 12.9 OE 30 12th EX 39 F-K 6.7 OE 30 12th EX 40 F-K 8.1 OE 30 12th EX 41 F-K 14.3 OE 30 12th EX 42 F-K 9.1 OE 30 12th EX 43 F-K 13.9 OE 30 12th EX 44 F-K 13.9 OE 30 12th EX 45 F-K 11.1 OE 30 12th EX 46 F-K 11.2 OE 30 12th EX 47 F-K 11.9 OE 30 12th EX 48 F-K 11.2 OE 30 12th EX 49 F-K 10.3 OE 30 12th EX 50 F-K 21.3 OE 30 12th EX Brown, Yamashiro, & Ogane (2001)

1 330 Intro level OE 30 11th EX, AC 2 330 Intro level OE 30 11th EX, AC

3 330 Intro level OE 30 11th EX, AC 4 330 Intro level OE 30 11th EX, AC 5 330 Intro level OE 30 11th EX, AC Chapelle (1988) 1 450 OE 50 7th AC Chapelle & Abraham (1990) 1 509 OE 35 11th 2 509 OE 35 Rational 3 509 MC 35 Rational EX Farhady & Keramati (1996) 1 337 OE 80 4th EX 2 337 OE 55 5th EX 3 337 OE 41 8th EX 4 337 OE 35 9th EX 5 337 OE 26 12th EX 6 337 OE 21 15th EX 7 337 OE 20 16th EX 8 337 OE 18 18th EX 9 337 OE 43 7th EX Flahive (1980) 1 400 OE 50 7th AC Hinofotis (1980) 1 427 OE 50 7th EX, AC


128

Hinofotis & Snow (1980) 1 427 MC & OE 50 7th EX, AC 2 427 MC & OE 50 7th EX, AC Ilyin, Spurling, & Seymour (1987) 1 393 OE 50 7th AC Irvine, Atai, & Oller (1974) 1 394 OE 50 7th Jafarpur (1995) 1 125 Flesch 79 OE 25 6th EX, AC Jafarpur (1996) 1 193 Flesch 79 OE 25 6th EX, AC Jonz (1976) 1 706 MC 33 EX Jonz (1987) 1 750 OE 50 15th 1 750 OE 50 15th 2 750 OE 50 Rational 2 750 OE 50 Rational Jonz (1991) 1 OE 50 7th AC 2 OE 50 7th AC 3 OE 50 7th AC 4 OE 50 7th AC Klein-Braley (1983) 1 OE 34 9th EX, AC 2 OE 46 7th EX, AC 3 OE 30 10th EX, AC 4 OE 43 8th EX, AC 5 OE 35 10th EX, AC 6 OE 50 6th EX, AC 7 OE 35 10th EX, AC 8 OE 50 6th EX, AC 9 OE 40 7th EX, AC 10 OE 30 10th EX, AC 11 OE 40 7th EX, AC 12 OE 30 10th EX, AC Kobayashi (2002) 1 357.8b Fry 8.4 OE 25 13th EX, AC 2 357.8 Fry 8.4 OE 25 13th EX, AC 3 357.8 Fry 8.4 OE 25 13th EX, AC 4 357.8 Fry 8.4 OE 25 13th EX, AC 5 380.5 Fry 8.2 OE 25 13th EX, AC 6 380.5 Fry 8.2 OE 25 13th EX, AC 7 380.5 Fry 8.2 OE 25 13th EX, AC 8 380.5 Fry 8.2 OE 25 13th EX, AC Mackay & Williamson (1979) 1 185 OE 24 7th Mullen (1979) 1 7th grade OE 50 10th EX, AC 2 12th grade OE 46 10th EX, AC Oller & Inal (1971) 1 OE 50 Rational AC Oller (1972) 1 375 Flesh 100 OE 7th EX, AC 2 375 Flesh 77 OE 7th EX, AC 3 375 Flesh 69 OE 7th EX, AC Pike (1979) 1 OE 25 10th EX, CE 2 OE 25 10th EX, CE 3 OE 25 10th EX, CE Wilson (1980) 1 200c OE 5th 2 200 OE Rational 3 200 OE 5th 4 200 OE Rational

b The readability value in Kobayashi’s (2002) study was an average of the four cloze test passages taken from the same text.

c Approximately 200 words.


129

REFERENCES

[Note. Studies that are included in the Meta-Analysis are indicated with an asterisk (*).]

*Abraham, R. G., & Chapelle, C. A. (1992). The meaning of cloze test scores: An item difficulty

perspective. Modern Language Journal, 76(4), 468-479.

Alderson, J. (1978). A study of the cloze procedure with native and non-native speakers of

English. Unpublished doctoral dissertation, University of Edinburgh, Edinburgh, UK.

Alderson, J. C. (1979a). Scoring procedures for use on cloze tests. In C. A. Yorio, K. Perkins, &

J. Schachter (Eds.), On TESOL '79 (pp. 193-205). Washington, DC: TESOL.

*Alderson, J. C. (1979b). The cloze procedure and proficiency in English as a foreign language.

TESOL Quarterly, 13(2), 219-227.

Alderson, J. (1980). Native and non-native speaker performance on cloze tests. Language

Learning, 30, 59-76.

ALTE. (1990). Multilingual glossary of language testing terms. Cambridge: Cambridge

University.

*Bachman, L. F. (1982). The trait structure of cloze test scores. TESOL Quarterly, 16, 61-70.

*Bachman, L. (1985). Performance on cloze tests with fixed-ratio and rational deletions. TESOL

Quarterly, 19, 535-555.

*Bensoussan, M., & Ramaz, R. (1984). Testing EFL reading comprehension using a multiple-

choice rational cloze. Modern Language Journal, 68(3), 230-239.

Bormuth, J. R. (1965). Validities of grammatical and semantic classifications of cloze test scores.

In J. A. Figurel (Ed.), Reading and inquiry (pp. 283-285). Newark, DE: International

Reading Associates.

Bormuth, J. (1967). Comparable cloze and multiple-choice comprehension tests scores. Journal

of Reading, 10, 291-299.

*Briere, E.,J., & Hinofotis, F.,B. (1979). Cloze test cutoff points for placing students in ESL

classes. In E. J. Briere & F. B. Hinofotis (Eds.), Concepts in language testing: Some recent

studies (pp. 12-20). Washington D.C.: TESOL.

Brown, J. D. (1978). Correlational study of four methods for scoring cloze tests. Unpublished

master’s thesis, University of California Los Angeles, CA.

*Brown, J. D. (1980). Relative merits of four methods for scoring cloze tests. Modern Language

Journal, 64, 311-317.


130

Brown, J. D. (1982). Language testing: Norm-referenced. Zhongshan Daxue English

Studies, 1(1), 192-204.

*Brown, J. D. (1983). A closer look at cloze: Validity and reliability. In J. W. Oller (Ed.), Issues

in language testing research (pp. 237-250). Rowley, MA: Newbury House.

Brown, J. D. (1984). A cloze is a cloze is a cloze? In J. Handscombe, R. A. Orem & B. P. Taylor

(Eds.), On TESOL '83 (pp. 109-119). Washington, DC: TESOL.

*Brown, J. D. (1988). Tailored cloze: Improved with classical item analysis techniques.

Language Testing, 5, 19-31.

Brown, J. D. (1989). Cloze item difficulty. JALT Journal, 11, 46-67.

*Brown, J. D. (1993). What are the characteristics of natural cloze tests? Language Testing, 10,

93-116.

Brown, J. D. (1994). A closer look at cloze: Validity and reliability. In J. Oller., & J. Jon (Eds.),

Cloze and coherence. Lewisburg, PA: Associated University Presses.

Brown, J. D. (1998). An EFL readability index. JALT Journal, 20(2), 7-36.

Brown, J. D. (2002). Do cloze tests work, or is it just an illusion? Second language studies:

Working papers of the Department of Second Language Studies, University of Hawaii, 21(1),

79-125.

Brown, J. D., Yamashiro, A., & Ogane, E. (1999). Tailored cloze: three ways to improve cloze

tests. University of Hawaii Working Papers in ESL, 17(2), 107-129.

*Brown, J. D., Yamashiro, A. D., & Ogane, E. (2001). The Emperor’s new cloze: Strategies for

revising cloze tests. In T. Hudson & J. D. Brown (Eds.), A focus on language test

development (pp. 143-161). Honolulu, HI: University of Hawai‘i Press.

*Chapelle, C. A. (1988) Field independence: a source of language test variance? Language

Testing, 5, 62-82.

*Chapelle, C. A., & Abraham. R. G. (1990). Cloze method: What differences does it make?

Language Testing, 7, 121-146.

Chavez-Oller, M. A., Chihara, T., Weaver, K. A., & Oller, J. W. (1985). When are cloze items

sensitive to constraints across sentences? Language Learning, 35, 181-206.

Chihara, T., Oller, J. W., Jr., Weaver, K. A., & Chavez-Oller, M. A. (1977).

Are cloze items sensitive to constraints across sentences? Language Learning, 27, 63-73.

Cohen, A. (1980). Testing language ability in the classroom. Rowley, MA: Newbury House.


131

Conrad, C. (1970). The cloze procedure as a measure of English proficiency. Unpublished

master’s thesis, University of California Los Angeles, Los Angeles, CA.

Crawford, A. (1970). The cloze procedure as a measure of reading comprehension of elementary

level Mexican-American and Anglo-American children. Unpublished doctoral dissertation,

University of California Los Angeles, Los Angeles, CA.

Darnell, D. (1970). Clozentropy: A procedure for testing English language proficiency of foreign

students. Speech Monographs, 37, 36-46.

Farhady, H. (1982). Measures of language proficiency from the learner's perspective. TESOL

Quarterly, 16, 43-59.

*Farhady, H., & Keramati, M. N. (1996). A text-driven method for the deletion procedure in

cloze passages. Language Testing, 13(2), 191-207.

*Flahive, D. E. (1980). Separating the g factor from reading comprehension. In J. W. Oller, Jr. &

K. Perkins (Eds.), Research in language testing (pp. 34-46). Rowley, MA: Newbury House.

Gallant, R. (1965). Use of cloze tests as a measure of readability in the primary grades. In J. A.

Figurel (Ed.), Reading and inquiry (pp. 286-287). Newark, DE: International Reading

Associates.

*Hinofotis, F. B. (1980). Cloze as an alternative method of ESL placement and proficiency

testing. In J. W. Oller, Jr. & K. Perkins (Eds.), Research in language testing (pp. 121-128).

Rowley, MA: Newbury House.

*Hinofotis, F. B., Snow, B. G. (1980). An alternative cloze testing procedure: Multiple-choice

format. In J. W. Oller, Jr. & K. Perkins (Eds.), Research in language testing (pp. 129-133).

Rowley, MA: Newbury House.

*Ilyin, D., Spurling, S., & Seymour, S. (1987). Do learner variables affect cloze correlations?

System, 15, 149-160.

*Irvine, P., Atai, P., & Oller, J. W. Jr. (1974). Cloze, dictation, and the Test of English as a

Foreign Language. Language Learning, 24, 245-252.

*Jafarpur, A. (1995). Is C-testing superior to cloze? Language Testing, 12(2), 194-216.

*Jafarpur, A. (1996). Native speaker performance validity: In vain or for gain? System, 24(1),

89-95.

*Jonz, J. (1976). Improving on the basic egg: The M-C cloze. Language Learning, 26, 255-256.


132

*Jonz, J. (1987). Textual cohesion and second language comprehension. Language Learning, 37,

409-438.

Jonz, J. (1990). Another turn in the conversation: What does cloze measure? TESOL Quarterly,

24(1), 61-83.

*Jonz, J. (1991). Cloze item types and second language comprehension. Language Testing, 8(1),

1-22.

*Klein-Braley, C. (1983). A cloze is a cloze is a question. In J. Oller (Ed.), Issues in language

testing research (pp. 218-228). Rowley, MA: Newbury House.

*Kobayashi, M. (2002). Cloze tests revisited: Exploring item characteristics with special

attention to scoring methods. Modern Language Journal, 86(4), 571-586.

Kuder, G. F., & Richardson, M. W. (1937). The theory of the estimation of test reliability.

Psychometrika, 2(3), 151-160.

*Mackay, R., & Williamson, B. (1979). A trial use of cloze test procedure in the assessment of

reading comprehension. Studia Anglica Posnaniensia, 10, 211-220.

Markham, P. L. (1985). The rational deletion cloze and global comprehension in German.

Language Learning, 35, 423-430.

*Mullen, K. (1979). More on cloze tests as tests of proficiency in English as a second language.

In E. J. Briere & F. B. Hinofotis (Eds.), Concepts in language testing: Some recent studies

(pp. 21-32). Washington, DC: TESOL.

Norris, J. M., & Ortega, L. (2006). Synthesizing research on language learning and teaching.

Philadelphia, PA: John Benjamins.

Oller, J. W. (1972a). Dictation as a test of ESL proficiency. In H. B. Allen & R. N. Campbell

(Eds.), Teaching English as a second language: A book of readings (pp. 346-354). New

York: McGraw-Hill.

*Oller, J. W. (1972b). Scoring methods and difficulty levels for cloze tests of proficiency in

English as a second language. Modern Language Journal, 56, 151-158.

Oller, J. (1979). Language tests at school: A pragmatic approach. London: Longman.

*Oller, J. W. Jr., & Inal, N. (1971). A cloze test of English prepositions. TESOL Quarterly, 5,

315-326.

Perkins, K., & Linnville, S. E. (1987). A construct definition study of a standardized ESL

vocabulary test. Language Testing, 4(2), 125-141.


133

*Pike, L. W. (1979). An evaluation of alternative formats for testing English as a second

language. TOEFL Research Report 2. Princeton, NJ: ETS.

Porter, D. (1983). The effect of quantity of context on the ability to make linguistic predictions:

A flaw in a measure of general proficiency. In A. Hughes & D. Porter (Eds.), Current

developments in language testing (pp. 63-74). London: Academic Press.

Revard, D. (1990). Tailoring the cloze to fit: Improvement of cloze tests through classical item

analysis. Unpublished scholarly paper. University of Hawaii at Manoa, Honolulu, HI.

Ruddell, R. (1964). A study of the cloze comprehension technique in relation to structurally

controlled reading material. Improvement of Reading Through Classroom Practice, 9,

298-303.

Stubbs, J. B., & Tucker, G. R. (1974). The cloze test as a measure of ESL proficiency for Arab

students. Modern Language Journal, 58, 239-241.

Taylor, W. L. (1953). Cloze procedure: a new tool for measuring readability. Journalism

Quarterly, 30, 414-438.

Thomas, M. (1994). Assessment of L2 proficiency in second language acquisition research.

Language Learning, 44, 307-336.

Thomas, M. (2006). Research synthesis and historiography: The case of assessment of second

language proficiency. In J. M. Norris, & L. Ortega (Eds.), Synthesizing research on language

learning and teaching (pp. 279-298). Philadelphia, PA: John Benjamins.

*Wilson, C. B. (1980). Can ESL cloze tests be contrastively biased?—Vietnamese as a test case.

In J. W. Oller, Jr. & K. Perkins (Eds.), Research in language testing (pp. 208-215). Rowley,

MA: Newbury House.

Date post:	07-Feb-2018
Category:	Documents
Upload:	trinhkien
View:	266 times
Download:	3 times

A META-ANALYSIS OF SECOND LANGUAGE CLOZE TESTING · PDF fileA META-ANALYSIS OF SECOND LANGUAGE...

Documents